Accounting & Audit
Use AI in Accounting and Audit Without Exposing Client Data
- 01
Identify sensitive values
- 02
Replace with stable tokens
- 03
On API traffic, restore inside your boundary
574
Built-in detectors for personal data and secrets
47
Country packs, checksum-validated where the format has one
9
AI providers behind one gateway, plus the desktop agent and proxy
Where client data meets AI in a firm
A tax senior has twelve client emails and a partly completed return, and asks a chatbot to list what's missing. An audit associate pastes a ledger extract and asks for unusual entries. A payroll specialist wants a formula fixed and pastes the sheet to show the problem, with every employee's salary and social security number still in it.
Accountants and auditors owe their clients a professional duty of confidentiality on top of data protection law. ICAEW's guidance on generative AI tells members not to load confidential information into public tools. Staff under deadline pressure will still look for help, and a block alone tends to move that work onto personal phones and accounts.
Maskflare keeps the work going while taking identifiers out: the desktop agent and browser protection for people, the gateway for the tools your firm builds, and Exact Data Match for your own client and engagement lists. Masking doesn't hide the numbers in a set of accounts, so the firm still needs a rule on what may go to which tool.
Where accounting & audit is actually exposed
5 failure modes specific to this environment, not a restatement of general security advice.
01
Client correspondence pasted for drafting
An email thread carries a client's name, contact details, tax number and bank details. The desktop agent masks them before the prompt reaches ChatGPT, Claude, Gemini, DeepSeek, Perplexity, Mistral Le Chat and Microsoft 365 Copilot. The draft may show tokens where it refers to them, and the person fills those in or keeps them.
02
Payroll files shared to fix a formula
A payroll sheet can hold names, national IDs, IBANs and salaries. CSV and text uploads are masked in place in the browser where they're readable in the request; an upload to a separate storage host, such as ChatGPT's current flow, is refused. Identifiers are detected; a salary figure on its own isn't, so block or warn on these files by policy.
03
Ledger and bank extracts
Extracts carry IBANs, card numbers and counterparty names. Checksum validation finds valid IBANs and card numbers without flagging every long reference number, and a dictionary or Exact Data Match covers client and supplier names.
04
Client names that aren't in any word list
Company names are hard to detect by pattern, and a list of who you act for is itself confidential. Upload client and engagement names to Exact Data Match, so they're masked from keyed hashes without being shared in clear text.
05
Unpublished financial results
Pre-announcement figures for a listed client are highly sensitive, and they aren't identifiers. Masking won't catch a revenue number, so use approved tools and block rules for this work, and treat the audit team's access as a separate question.
Who this page is for
- Managing partner
- Wants productivity from AI without one paste becoming a confidentiality breach with a client.
- Head of audit quality
- Needs to know what left the firm, and to show that use of AI on engagements is controlled.
- Tax and payroll leads
- Want staff to get drafting and formula help without personal and tax data reaching a provider.
- Data protection officer
- Has to document which AI providers receive personal data, and on what basis.
- IT lead at a mid-sized firm
- Needs something deployable through existing device management, without a large security team to run it.
How a team would actually run this
Using AI on client correspondence
- 01 Deploy the desktop agent with Intune, Jamf or Group Policy. Only AI hosts are routed through Maskflare, so other traffic isn't touched.
- 02 Apply the GDPR profile, or the US PII profile for US clients, and turn on the country packs for your clients' markets.
- 03 Upload client and engagement names to Exact Data Match with `maskflare edm upload`, from your practice management system.
- 04 Staff work as usual. The provider sees tokens. The draft may show tokens where it refers to masked names and numbers, and staff fill them in or keep them.
Setting rules for sensitive engagements
- 01 Start in alert mode for two weeks to see which categories turn up, in which AI apps.
- 02 Move everyday identifiers to mask, and set a warn rule with justification where a partner wants a record of the reason.
- 03 Block credentials and any dictionary terms tied to listed clients before their results are public.
- 04 Export the audit log and coverage report for your quality file, and require a second approver for rule changes.
Which modules apply, and why
Each links to the module page, where the boundaries and development status are set out in full. Or start at the Maskflare platform overview.
Detection engine
Maskflare Detection Engine
Detects and masks names, tax and VAT numbers, IBANs, cards and national IDs before a prompt reaches an AI provider, and on API traffic restores them in the answer.
Detection engine
Maskflare Exact Data Match
Masks your client and engagement names from keyed hashes, so the list of who you act for isn't shared in clear text.
For your people
Maskflare Workforce
The desktop agent covers staff on AI chat sites and is deployed through device management.
For your people
Maskflare User Coaching
A warn rule asks staff to give a reason before sending a sensitive category, which gives a partner a record to review.
For your people
Maskflare Browser Protection
Masks text typed into AI chat sites and CSV or text uploads where they're readable in the request. The answer isn't rewritten.
Engine and control
Maskflare Admin Console
Holds rules, four-eyes approval, the audit log and coverage reports for your quality file.
The data, and how it's found
Typical data in this sector, the detectors that cover it, and where you'll need your own rules.
| Data | How Maskflare finds it | Profile or rule |
|---|---|---|
| Tax numbers, such as German Steuer-ID | 47 country packs with checksum and format validation | GDPR, US PII |
| VAT and company IDs, such as Swiss UID | EU VAT and country pack detectors | GDPR |
| IBANs and bank details | IBAN validation by checksum | GDPR, UK GDPR |
| Payment cards | Card detection with the Luhn checksum | PCI DSS v4.0 |
| Employee names and national IDs | Person names pack and country packs | GDPR, US PII |
| Client and engagement names | Exact Data Match or a custom dictionary | Your own rules |
| Salaries and financial figures | No built-in detector: use custom rules, warn or block | Your policy decision |
Regulatory context
Readiness and relevance, not certification. Nothing here is a claim to hold an audit or authorization we do not have.
- IESBA Code, Subsection 114
- Sets the confidentiality principle for professional accountants. Masking is a safeguard you can document under it, and it doesn't settle the question by itself.
- ICAEW generative AI guidance
- Tells members that respecting confidentiality means not loading confidential information into public generative AI tools. Check what your own institute says, and whether a masked prompt meets it.
- EU Audit Directive 2006/43/EC
- Sets confidentiality and professional secrecy duties for statutory auditors. National implementations differ, and your firm's reading of them applies.
- Swiss Code of Obligations Art. 730b and Criminal Code Art. 321
- Set confidentiality duties for auditors in Switzerland, with professional secrecy backed by criminal law. Hosting and on-premises options can help, but legal assessment stays with you.
- GDPR and revised Swiss FADP
- Apply to personal data in client files and payroll. Reversible tokens remain pseudonymized personal data for you, so your legal basis still applies.
What Maskflare does not do
- Masking removes identifiers, not numbers. A set of accounts with the name removed can still identify a company to someone who knows it.
- Salaries, revenue and other figures have no built-in detector. Control them with custom rules, warn or block, or by choosing approved tools.
- It doesn't check the AI's work. Staff remain responsible for professional judgment and for documenting what they relied on.
- Documents, including scanned pages read with OCR, are scanned under your document policy and recorded or refused, not rewritten. In browser apps, picture uploads pass unread or are refused, and uploads that go to a separate storage host are refused.
- SOC 2 Type II and ISO 27001 audits are in progress.
5 questions to ask any vendor here
Including us. If our answer is worse than someone else's, you should know that before you buy, not after.
- 01Does the provider ever receive a client name, or only tokens?
- 02Can it match our client list without us handing that list over in clear text?
- 03What happens to confidential figures that aren't identifiers, such as unpublished results?
- 04What record do we get of who sent which category, to which app, and why?
- 05Can we deploy it through our existing device management without a dedicated security team?
Workflow guides for this industry
How a team applies each of these, including the contrast with the tool it replaces.
Definitions worth agreeing on first
Accounting & Audit questions we get asked
Can accountants use ChatGPT with client data?
That depends on your professional body, engagement terms and data protection position, so check them first. With the desktop agent installed, identifiers are masked before they reach ChatGPT, which supports that decision without making it for you. This is not legal advice.
Does masking satisfy our confidentiality duty?
No single tool does. Masking is a safeguard you can document, and it reduces what a provider receives. Confidential facts that aren't identifiers can still be sent, so you need rules for those too.
How do you protect client names?
Upload them to Exact Data Match, where only keyed hashes leave your environment, or add a custom dictionary. Company names are hard to find by pattern alone, so a list works better.
Does Maskflare detect salaries or revenue figures?
No. A number with no fixed format isn't something it can find reliably. Use a custom rule for labelled fields, a warn or block rule on those files, or a policy on approved tools.
Is this suitable for a smaller firm?
Yes, if you have device management or someone to run it. Most teams start with 10 to 50 devices and one enrollment token, in alert mode.
Where is Maskflare hosted?
In the EU by default, with on-premises deployment available for enterprise. Prompts and responses are never stored; masked values are kept encrypted for 30 days so answers can be restored.
See it on your own data
Start a free trial or talk to us with the identifiers and AI workflows your accounting & audit team actually handles. Bring the case that makes it hard, because that is the part worth testing.
See it on your own data
Try Maskflare free.
Bring your hardest prompts.
28-day free trial or 10,000 requests, whichever comes first. Point one app at the gateway or enroll a few laptops,
and watch what gets masked. Or talk to us first.
- SOC 2 Type II: In progress
- ISO 27001: In progress
- EU hosting by default · on-premises available
- Prompts and files never stored
- Never used to train AI models
- GDPR · nDSG · DPA available
- Trust center →